A method and system for purifying and recycling tail water of aquaculture by hierarchical circulation

Abstract

The application discloses a kind of hierarchical circulation's aquaculture tail water purification reuse method, applied to circulating water aquaculture system, comprising: construct the full-dimension behavior fingerprint library of covering single factor, double factor complex and multi-factor coupling stress, build the coupling model of feeding metabolism, behavior response, tail water pollution load and multi-stress factor characteristic decoupling algorithm, lightweight deployment cloud and edge gateway;Collect aquaculture biological behavior and water quality data, with living stress microbehavior as the only basis for decision, through pseudo stress filtering, stress decoupling identification, working condition determination and target verification output control instruction and closed loop iteration;Performing feeding and oxygenation collaborative control and tail water hierarchical purification reuse.The scheme can accurately identify complex water stress, avoid sensor distortion and response lag problem, with risk prevention and control front, control precision, source pollution reduction, reuse safety controllable characteristics.

Description

Technical Field

[0001] This invention relates to the field of aquaculture environmental protection and intelligent management technology, specifically to a method and system for purifying and reusing aquaculture wastewater through a graded recycling system. Background Technology

[0002] With the rapid development of intensive and high-density aquaculture, the control of pollutant discharge from aquaculture wastewater is becoming increasingly stringent, and the shortage of aquaculture water resources is becoming more prominent. Therefore, there is a need to develop efficient, precise, and safe technologies for purifying and reusing aquaculture wastewater. Existing aquaculture wastewater purification and reuse technologies have two major shortcomings: It can only identify single water quality stresses and cannot decouple and accurately determine the characteristics of dual-factor and multi-factor complex water quality stresses. There are blind spots in stress identification and the accuracy of water quality risk prevention and control is low. It uses water quality sensor data as the sole basis for control and reuse safety judgment. The sensors are prone to drift and distortion, and the response of physicochemical indicators lags behind biological stress. The source pollution reduction effect is poor and the reuse safety guarantee capability is insufficient.

[0003] To address the aforementioned technical challenges, this invention proposes a technical solution that uses the subclinical microbehavior of live cultured organisms as the sole core decision-making basis, constructs a multi-dimensional behavioral fingerprint database with decoupled multiple stress factors, and combines an edge-cloud isomorphic architecture with intelligent collaborative management to achieve graded purification of wastewater, source pollution reduction, and safe reuse. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for purifying and reusing aquaculture wastewater in a graded and cyclical manner, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for purifying and reusing aquaculture wastewater through a graded recycling system, comprising: A comprehensive behavioral fingerprint database of the target aquaculture species is constructed. The database includes specific micro-behavioral features corresponding to single-factor water quality stress, dual-factor composite stress, and multi-factor coupled stress, clarifying the one-to-one correspondence between different stress combinations and the subclinical micro-behaviors of aquaculture organisms. Simultaneously, a dynamic model coupling the feeding metabolism, behavioral response, and wastewater pollution load of aquaculture organisms is constructed, along with a multi-stress factor feature decoupling algorithm based on a temporal feature separation network using an attention mechanism. The comprehensive behavioral fingerprint database, coupled dynamic model, and multi-stress factor feature decoupling algorithm are then lightweighted and compressed, and simultaneously deployed to the cloud and edge gateways to build an edge-cloud isomorphic infrastructure. Collect core data on the live behavior of aquaculture organisms and auxiliary data on water quality, and transmit them to the edge gateway and the cloud; Based on the aforementioned full-dimensional behavioral fingerprint database, and using the micro-behaviors of aquaculture organisms under stress as the sole core basis for determining operating conditions, triggering control measures, and verifying authenticity, the system first filters out pseudo-stresses caused by non-water quality stress. Then, through the aforementioned multi-stress factor feature decoupling algorithm, it identifies the features of single-factor and compound stresses, the type of inducing factors, the combination of stresses, and the intensity of stresses in real stress behavior data. Subsequently, based on the stress identification results, it completes the standardization determination of operating conditions across the entire gradient. For operating conditions that trigger verification, it completes the verification through targeted single-variable intervention and changes in biological behavioral responses, outputs control decision instructions, and simultaneously updates the full-dimensional behavioral fingerprint database, the multi-stress factor feature decoupling algorithm, and the coupled dynamic model to complete the closed-loop iteration. Based on the control decision instructions and stress identification results, feed-oxygenation coordinated control is implemented to reduce the pollution load of tailwater from the source; based on the behavioral characteristics of aquaculture organisms, stress identification results and the prediction results of coupled dynamic models, tailwater graded purification and recycling control is implemented.

[0006] According to the above technical solution, the process of filtering non-water quality stress pseudo-stress includes: identifying real-time collected live behavior data of cultured organisms based on pre-stored water quality stress-specific micro-behavioral features and non-water quality stress pseudo-stress behavioral features in a full-dimensional behavioral fingerprint database; extracting the change type, duration, and trend of behavioral features; linking water temperature monitoring data, on-site operation records, culture density data, and disease auxiliary monitoring data to match the triggering scenarios and associated causes of behavioral abnormalities; filtering non-water quality stress pseudo-stress behaviors caused by human operation disturbance, environmental noise interference, sudden water temperature changes, culture density stress, and biological diseases, and excluding behavioral abnormality data caused by non-water quality factors; locking in real stress behaviors that are only caused by water quality stress and have progressive change characteristics, and outputting the filtered biological behavioral feature data to the stress decoupling and decision-making stage.

[0007] According to the above technical solution, the multi-stress factor feature decoupling algorithm of the attention mechanism-based temporal feature separation network includes: S1 Input Layer: The time-series data of the micro-behavior of the cultured organisms after pseudo-stress filtering are divided into four types of time-series feature sequences according to the behavioral dimension: respiratory rhythm, swimming characteristics, feeding response, and stress avoidance. After normalization, the data are output to the time-series feature encoding layer. S2 Temporal Feature Encoding Layer: A bidirectional long short-term memory network Bi-LSTM is used to perform contextual encoding on the input multi-dimensional temporal feature sequence, capturing the long temporal dependencies and progressive change features of different behavioral features, and outputting a high-dimensional behavioral feature map that integrates temporal associations; S3 Multi-Scale Attention Separation Layer: A parallel multi-scale channel attention module and a temporal attention module are constructed. The channel attention module learns feature weights corresponding to different water quality stress factors for feature channels of different behavioral dimensions, strengthening stress-related specific feature channels and suppressing irrelevant features. The temporal attention module captures the temporal variation patterns of behavioral features corresponding to different stress factors, locking in key temporal segments of stress features. Through a dual attention mechanism, a weighted output is obtained to obtain decoupled feature subsets corresponding to different stress factors. S4 Decoupling Feature Matching Layer: The decoupled stress feature subsets are matched with the pre-stored single-factor, dual-factor, and multi-factor stress-specific behavioral fingerprints in the full-dimensional behavioral fingerprint database. By using a preset feature similarity threshold, the types and combinations of single-factor stress, dual-factor composite stress, and multi-factor coupled stress are distinguished, and one or more water quality stress triggers that induce stress behavior are accurately located and the corresponding stress intensity is quantified. S5 Output Layer: Outputs the types and combinations of water quality stress factors identified, as well as the corresponding stress intensity, and simultaneously outputs the specific behavioral characteristic data corresponding to each stress factor for subsequent working condition judgment and control.

[0008] According to the above technical solution, the standardized determination of the full gradient operating condition includes: Pre-set biological behavioral stress grading thresholds for the entire growth cycle of the target aquaculture species, and for single-factor and compound stresses. These thresholds include a slight change threshold corresponding to subclinical mild stress and a stress grading warning threshold corresponding to clear stress abnormalities. The behavioral characteristics of the aquaculture organisms after pseudo-stress filtering and stress decoupling are compared with the stress intensity. The entire process uses biological behavioral characteristics as the sole criterion for judgment, sequentially completing the standardized determination of four types of working conditions: Stress-free steady-state condition: All behavioral characteristics of aquaculture organisms are within the stress-free baseline range. Regardless of fluctuations in water quality auxiliary data, the condition is considered steady-state and no control actions are triggered. Stress-abnormal condition: Behavioral characteristics of aquaculture organisms reach the stress grading threshold. Regardless of whether water quality auxiliary data meets the standards, control actions corresponding to the identified stress triggers are triggered, and the live organism verification process is initiated. Data contradiction condition: Water quality auxiliary data shows anomalies, but the behavioral characteristics of aquaculture organisms remain stable within the stress-free baseline range. The decision-making authority for water quality sensor data is directly locked, and the live organism verification process is initiated. The biological behavioral characteristics are used as the sole criterion for judgment, and control trigger commands corresponding to the three types of conditions are output.

[0009] According to the above technical solution, the verification is completed through targeted intervention and changes in biological behavioral responses, including: Based on the standardized judgment results of the working conditions, for the clearly defined abnormal working conditions that trigger verification, targeted intervention actions are performed on the corresponding water quality factors based on the combination and intensity of water quality stress causes identified by decoupling, and the remaining water quality factors and aquaculture environment parameters are kept stable throughout the process; for the data contradiction working conditions that trigger verification, targeted verification intervention is performed on the corresponding water quality sensors. The entire process involves synchronously collecting time-series data on the microbehavioral behavior of aquaculture organisms before, during, and after the intervention. Data on the magnitude of changes in behavioral characteristics, the rate of mitigation, and the response changes in the duration of these changes are extracted. Using the behavioral response changes of aquaculture organisms as the sole verification benchmark, the changes in behavioral characteristics before and after the intervention are compared to distinguish between actual water quality anomalies and sensor data anomalies. Based on the differentiation and judgment results, corresponding actions are taken: For results determined to be true water quality anomalies, emergency control measures matching the stress type and intensity are implemented, and the characteristics of the abnormal behavior and intervention response data are simultaneously added to the full-dimensional behavioral fingerprint database; for results determined to be sensor drift distortion, the time point at which the stress behavior of aquatic organisms completely returns to the stress-free baseline range after targeted intervention is identified, and the online correction of sensor parameters is completed based on the measured data collected synchronously at that point; for results determined to be sensor fault jumps, equipment self-checks and fault alarms are triggered, and the control decision-making authority for abnormal data is locked.

[0010] According to the above technical solution, the step of executing a coordinated feeding-oxygenation control based on control decision instructions and stress identification results to reduce effluent pollution load from the source includes: Based on the coupled dynamic model, combined with the real-time identified feeding behavior characteristics and metabolic stress microbehavioral characteristics of farmed organisms, the nutritional requirements and pollutant production patterns of the target farmed species throughout their entire growth cycle are matched, and the feeding plan and core feeding parameters are dynamically adjusted to reduce the generation of uneaten feed and fecal pollutants from the source. Based on the dissolved oxygen stress-specific micro-behavioral features pre-stored in the full-dimensional behavioral fingerprint database, combined with the real-time identified dissolved oxygen stress behavior features of aquaculture organisms and the metabolic oxygen consumption change trend output by the coupled dynamic model, the trend of dissolved oxygen change in water body is predicted, multi-level predictive gradient oxygenation actions are executed, and the operating parameters of oxygenation equipment are dynamically adjusted.

[0011] According to the above technical solution, the step of implementing graded purification and recycling management of wastewater based on the behavioral characteristics of aquaculture organisms, stress identification results, and prediction results of a coupled dynamic model includes: Based on the coupled dynamic model, combined with the real-time identified feeding behavior characteristics and metabolic stress-related microbehavioral characteristics of aquaculture organisms, the pollutant load range, main pollutant types and output patterns of aquaculture wastewater are predicted, and the wastewater pollution load is divided into three levels: light, medium and heavy; and wastewater purification paths corresponding to different pollution load levels are matched. During the process of injecting purified recycled water into aquaculture, time-series data of the live microbehavior of aquaculture organisms are collected simultaneously. The baseline features and stress-specific features in the full-dimensional behavioral fingerprint database are compared. The safety of the recycled water for aquaculture is determined by using the live behavioral response of aquaculture organisms as the sole benchmark, and the reuse ratio, injection rate and reuse application scenarios of recycled water are dynamically adjusted.

[0012] Based on the above technical solution, this application also proposes a graded recycling system for purifying and reusing aquaculture wastewater, the system comprising: Edge-cloud architecture and model building module: This module is used to construct a full-dimensional behavioral fingerprint database of the target aquaculture species. The database includes specific micro-behavioral features corresponding to single-factor water quality stress, dual-factor composite stress, and multi-factor coupled stress, clarifying the one-to-one correspondence between different stress combinations and the subclinical micro-behaviors of aquaculture organisms. Simultaneously, a dynamic model coupling the feeding metabolism, behavioral response, and wastewater pollution load of aquaculture organisms is constructed, along with a multi-stress factor feature decoupling algorithm based on a temporal feature separation network using an attention mechanism. The full-dimensional behavioral fingerprint database, coupled dynamic model, and multi-stress factor feature decoupling algorithm are then lightweighted and compressed, and simultaneously deployed to the cloud and edge gateways to build an edge-cloud isomorphic infrastructure. Multi-source data acquisition and transmission module: used to collect core data on the behavior of aquaculture organisms and auxiliary water quality data, and transmit them to the edge gateway and the cloud; The intelligent decision verification module connects the edge-cloud architecture, model building module, and multi-source data acquisition and transmission module. Based on the full-dimensional behavioral fingerprint database, it uses the micro-behaviors of aquaculture organisms under stress as the sole core basis for determining operating conditions, triggering control measures, and verifying authenticity. First, it filters out pseudo-stresses caused by non-water quality stress. Then, through the multi-stress factor feature decoupling algorithm, it identifies single-factor and compound stress features, stress inducing factors, combination methods, and stress intensity in real stress behavior data. Subsequently, based on the stress identification results, it completes full-gradient standardization judgment of operating conditions. For operating conditions triggering verification, it completes verification through targeted single-variable intervention and changes in biological behavioral responses, outputs control decision instructions, and synchronously updates the full-dimensional behavioral fingerprint database, the multi-stress factor feature decoupling algorithm, and the coupled dynamic model to complete closed-loop iteration. Feeding-oxygenation coordinated control module: Connected to the intelligent decision-making and closed-loop iteration module, it is used to execute feeding-oxygenation coordinated control based on control decision instructions and stress identification results, so as to reduce the pollution load of tailwater from the source; Wastewater grading, purification, and reuse module: Connects the intelligent decision-making and closed-loop iteration module, the edge-cloud architecture and model building module, and is used to perform wastewater grading, purification, and recycling management based on the behavioral characteristics of aquaculture organisms, stress identification results, and the prediction results of coupled dynamic models.

[0013] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention constructs a full-dimensional behavioral fingerprint database covering single-factor, dual-factor composite, and multi-factor coupled stress, and combines it with a temporal feature separation network based on an attention mechanism to decouple the features of multiple stress factors. This enables accurate identification of composite water quality stresses, fills blind spots in stress identification, and significantly improves the accuracy of water quality risk prevention and control. Using the subclinical stress microbehavior of aquaculture organisms as the sole core decision-making basis for the entire process, it effectively avoids the defects of sensor drift distortion and lag in physicochemical indicator response, achieving accurate and reliable pre-emptive prevention and control of water quality risks. Relying on an edge-cloud isomorphic architecture, it completes the lightweight deployment and closed-loop iteration of models and algorithms. Through feeding-oxygenation synergistic management, it significantly reduces the pollution load of tailwater from the source. At the same time, it combines stress identification results to match graded purification paths and uses biological behavioral responses to determine the safety of reuse, effectively improving the energy efficiency of tailwater treatment and the safety and controllability of water resource reuse, ultimately achieving the goal of pollution reduction and efficiency improvement, and safe recycling in aquaculture. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a graded recycling method for purifying and reusing aquaculture wastewater, provided by an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see Figure 1 The flowchart below illustrates a method for purifying and reusing aquaculture wastewater through a graded recycling system, as provided in an embodiment of the present invention. Figure 1 It can be seen that the aforementioned method for purifying and reusing aquaculture wastewater through a tiered recycling system includes: A comprehensive behavioral fingerprint database of the target aquaculture species is constructed. The database includes specific micro-behavioral features corresponding to single-factor water quality stress, dual-factor composite stress, and multi-factor coupled stress, clarifying the one-to-one correspondence between different stress combinations and the subclinical micro-behaviors of aquaculture organisms. Simultaneously, a dynamic model coupling the feeding metabolism, behavioral response, and wastewater pollution load of aquaculture organisms is constructed, along with a multi-stress factor feature decoupling algorithm based on a temporal feature separation network using an attention mechanism. The comprehensive behavioral fingerprint database, coupled dynamic model, and multi-stress factor feature decoupling algorithm are then lightweighted and compressed, and simultaneously deployed to the cloud and edge gateways to build an edge-cloud isomorphic infrastructure. Collect core data on the live behavior of aquaculture organisms and auxiliary data on water quality, and transmit them to the edge gateway and the cloud; Based on the aforementioned full-dimensional behavioral fingerprint database, and using the micro-behaviors of aquaculture organisms under stress as the sole core basis for determining operating conditions, triggering control measures, and verifying authenticity, the system first filters out pseudo-stresses caused by non-water quality stress. Then, through the aforementioned multi-stress factor feature decoupling algorithm, it identifies the features of single-factor and compound stresses, the type of inducing factors, the combination of stresses, and the intensity of stresses in real stress behavior data. Subsequently, based on the stress identification results, it completes the standardization determination of operating conditions across the entire gradient. For operating conditions that trigger verification, it completes the verification through targeted single-variable intervention and changes in biological behavioral responses, outputs control decision instructions, and simultaneously updates the full-dimensional behavioral fingerprint database, the multi-stress factor feature decoupling algorithm, and the coupled dynamic model to complete the closed-loop iteration. Based on the control decision instructions and stress identification results, feed-oxygenation coordinated control is implemented to reduce the pollution load of tailwater from the source; based on the behavioral characteristics of aquaculture organisms, stress identification results and the prediction results of coupled dynamic models, tailwater graded purification and recycling control is implemented.

[0017] According to the above technical solution, the process of filtering non-water quality stress pseudo-stress includes: identifying real-time collected live behavior data of cultured organisms based on pre-stored water quality stress-specific micro-behavioral features and non-water quality stress pseudo-stress behavioral features in a full-dimensional behavioral fingerprint database; extracting the change type, duration, and trend of behavioral features; linking water temperature monitoring data, on-site operation records, culture density data, and disease auxiliary monitoring data to match the triggering scenarios and associated causes of behavioral abnormalities; filtering non-water quality stress pseudo-stress behaviors caused by human operation disturbance, environmental noise interference, sudden water temperature changes, culture density stress, and biological diseases, and excluding behavioral abnormality data caused by non-water quality factors; locking in real stress behaviors that are only caused by water quality stress and have progressive change characteristics, and outputting the filtered biological behavioral feature data to the stress decoupling and decision-making stage.

[0018] According to the above technical solution, the multi-stress factor feature decoupling algorithm of the attention mechanism-based temporal feature separation network includes: S1 Input Layer: The time-series data of the micro-behavior of the cultured organisms after pseudo-stress filtering are divided into four types of time-series feature sequences according to the behavioral dimension: respiratory rhythm, swimming characteristics, feeding response, and stress avoidance. After normalization, the data are output to the time-series feature encoding layer. S2 Temporal Feature Encoding Layer: A bidirectional long short-term memory network Bi-LSTM is used to perform contextual encoding on the input multi-dimensional temporal feature sequence, capturing the long temporal dependencies and progressive change features of different behavioral features, and outputting a high-dimensional behavioral feature map that integrates temporal associations; S3 Multi-Scale Attention Separation Layer: A parallel multi-scale channel attention module and a temporal attention module are constructed. The channel attention module learns feature weights corresponding to different water quality stress factors for feature channels of different behavioral dimensions, strengthening stress-related specific feature channels and suppressing irrelevant features. The temporal attention module captures the temporal variation patterns of behavioral features corresponding to different stress factors, locking in key temporal segments of stress features. Through a dual attention mechanism, a weighted output is obtained to obtain decoupled feature subsets corresponding to different stress factors. S4 Decoupling Feature Matching Layer: The decoupled stress feature subsets are matched with the pre-stored single-factor, dual-factor, and multi-factor stress-specific behavioral fingerprints in the full-dimensional behavioral fingerprint database. By using a preset feature similarity threshold, the types and combinations of single-factor stress, dual-factor composite stress, and multi-factor coupled stress are distinguished, and one or more water quality stress triggers that induce stress behavior are accurately located and the corresponding stress intensity is quantified. S5 Output Layer: Outputs the types and combinations of water quality stress factors identified, as well as the corresponding stress intensity, and simultaneously outputs the specific behavioral characteristic data corresponding to each stress factor for subsequent working condition judgment and control.

[0019] For example, AI high-definition behavioral cameras are deployed at the aquaculture site to continuously collect live micro-behavioral videos of Litopenaeus vannamei at a sampling frequency of 1Hz. After pseudo-stress filtering, non-water quality stress behavior data such as human disturbance and sudden changes in water temperature are removed, and time-series data of micro-behaviors related to water quality stress are retained for 10 minutes. In line with the physiological characteristics of Litopenaeus vannamei, the micro-behaviors are divided into four types of scene-based features: respiratory rhythm, swimming characteristics, feeding response, and stress avoidance. The algorithm is lightweight and adapted to the computing power of the edge gateway at the aquaculture site. The number of nodes in the Bi-LSTM hidden layer is set to 128, the feature similarity threshold is set to 0.85, and the stress intensity quantification range is 0~1, where 0~0.6 is mild, 0.6~0.8 is moderate, and 0.8~1 is severe.

[0020] The specific implementation steps of the algorithm are as follows: S1 Input Layer: The 10-minute micro-behavioral time-series data of Litopenaeus vannamei after pseudo-stress filtering is decomposed into independent time-series feature sequences according to the four types of scenario-based behavior dimensions mentioned above. Min-max normalization is used to map all feature values ​​to the 0~1 interval. After eliminating the difference in dimensions, the data is output to the time-series feature encoding layer. S2 Time-Series Feature Encoding Layer: A bidirectional long short-term memory network (Bi-LSTM) is used to perform context encoding on the four types of normalized time-series feature sequences. The focus is on capturing the long-term temporal dependencies of Litopenaeus vannamei under water quality stress, such as: the gill fanning frequency slowly increases under ammonia nitrogen stress and the swimming speed continuously decreases under low dissolved oxygen stress. The output is a 4-channel high-dimensional behavioral feature map with fused temporal correlation. S3 Multi-Scale Attention Separation Layer: A parallel multi-scale channel attention module and a time attention module are built. The channel attention module assigns scenario-based weights to the four types of behavioral feature channels. For ammonia nitrogen + nitrite stress, the weights of the respiratory rhythm channel are strengthened, and for low dissolved oxygen + high pH stress, the weights of the channel attention module are strengthened. The stress-enhancing swimming feature channel weights suppress interference from irrelevant features; the time attention module locks key temporal segments of stress features, such as feeding response periods and high oxygen consumption periods at night; through weighted fusion using a dual attention mechanism, it outputs independent decoupled feature subsets corresponding to each stress factor, including ammonia nitrogen, nitrite, low dissolved oxygen, and high pH; the S4 decoupled feature matching layer matches the decoupled feature subsets of each stress factor with the pre-stored single-factor and dual-factor composite stress-specific behavioral fingerprints of Litopenaeus vannamei in the full-dimensional behavioral fingerprint database using cosine similarity; when the similarity is ≥0.85, the match is considered successful, distinguishing between single-factor stress and dual-factor composite stress types, and quantifying the stress intensity based on the degree of feature deviation; the S5 output layer outputs scenario-based judgment results, for example: composite water quality stress: ammonia nitrogen + nitrite, stress intensity 0.87, simultaneously outputting the respiratory rhythm and swimming feature-specific behavioral data corresponding to this stress combination, directly transmitting them to subsequent operating condition judgment and tailwater management stages, providing data support for targeted intervention and graded purification.

[0021] This embodiment deeply binds the algorithm's input data, feature segmentation, network parameters, and matching rules with the Litopenaeus vannamei farming scenario and core water quality stress types. Those skilled in the art only need to adjust the behavioral feature definition, stress factor range, and algorithm parameters according to the target farming species to reproduce this multi-stress factor decoupling algorithm and achieve accurate identification of complex water quality stresses.

[0022] According to the above technical solution, the standardized determination of the full gradient operating condition includes: Pre-set biological behavioral stress grading thresholds for the entire growth cycle of the target aquaculture species, and for single-factor and compound stresses. These thresholds include a slight change threshold corresponding to subclinical mild stress and a stress grading warning threshold corresponding to clear stress abnormalities. The behavioral characteristics of the aquaculture organisms after pseudo-stress filtering and stress decoupling are compared with the stress intensity. The entire process uses biological behavioral characteristics as the sole criterion for judgment, sequentially completing the standardized determination of four types of working conditions: Stress-free steady-state condition: All behavioral characteristics of aquaculture organisms are within the stress-free baseline range. Regardless of fluctuations in water quality auxiliary data, the condition is considered steady-state and no control actions are triggered. Stress-abnormal condition: Behavioral characteristics of aquaculture organisms reach the stress grading threshold. Regardless of whether water quality auxiliary data meets the standards, control actions corresponding to the identified stress triggers are triggered, and the live organism verification process is initiated. Data contradiction condition: Water quality auxiliary data shows anomalies, but the behavioral characteristics of aquaculture organisms remain stable within the stress-free baseline range. The decision-making authority for water quality sensor data is directly locked, and the live organism verification process is initiated. The biological behavioral characteristics are used as the sole criterion for judgment, and control trigger commands corresponding to the three types of conditions are output.

[0023] According to the above technical solution, the verification is completed through targeted intervention and changes in biological behavioral responses, including: Based on the standardized judgment results of the working conditions, for the clearly defined abnormal working conditions that trigger verification, targeted intervention actions are performed on the corresponding water quality factors based on the combination and intensity of water quality stress causes identified by decoupling, and the remaining water quality factors and aquaculture environment parameters are kept stable throughout the process; for the data contradiction working conditions that trigger verification, targeted verification intervention is performed on the corresponding water quality sensors. The entire process involves synchronously collecting time-series data on the microbehavioral behavior of aquaculture organisms before, during, and after the intervention. Data on the magnitude of changes in behavioral characteristics, the rate of mitigation, and the response changes in the duration of these changes are extracted. Using the behavioral response changes of aquaculture organisms as the sole verification benchmark, the changes in behavioral characteristics before and after the intervention are compared to distinguish between actual water quality anomalies and sensor data anomalies. Based on the differentiation and judgment results, corresponding actions are taken: For results determined to be true water quality anomalies, emergency control measures matching the stress type and intensity are implemented, and the characteristics of the abnormal behavior and intervention response data are simultaneously added to the full-dimensional behavioral fingerprint database; for results determined to be sensor drift distortion, the time point at which the stress behavior of aquatic organisms completely returns to the stress-free baseline range after targeted intervention is identified, and the online correction of sensor parameters is completed based on the measured data collected synchronously at that point; for results determined to be sensor fault jumps, equipment self-checks and fault alarms are triggered, and the control decision-making authority for abnormal data is locked.

[0024] According to the above technical solution, the step of executing a coordinated feeding-oxygenation control based on control decision instructions and stress identification results to reduce effluent pollution load from the source includes: Based on the coupled dynamic model, combined with the real-time identified feeding behavior characteristics and metabolic stress microbehavioral characteristics of farmed organisms, the nutritional requirements and pollutant production patterns of the target farmed species throughout their entire growth cycle are matched, and the feeding plan and core feeding parameters are dynamically adjusted to reduce the generation of uneaten feed and fecal pollutants from the source. Based on the dissolved oxygen stress-specific micro-behavioral features pre-stored in the full-dimensional behavioral fingerprint database, combined with the real-time identified dissolved oxygen stress behavior features of aquaculture organisms and the metabolic oxygen consumption change trend output by the coupled dynamic model, the trend of dissolved oxygen change in water body is predicted, multi-level predictive gradient oxygenation actions are executed, and the operating parameters of oxygenation equipment are dynamically adjusted.

[0025] Preferably, the coupled dynamic model of the present invention is a hierarchical time-series prediction model, which consists of four modules linearly connected: biomarker acquisition, pollutant quantification, growth stage adaptation, and trend prediction. The training set is constructed using measured data of the entire growth cycle of Pacific salmon, and training is completed using 1D-CNN + attention mechanism and LSTM network. Core parameters and error judgment criteria are set. The model input is micro-behavioral data of feeding and metabolic stress of farmed organisms, and the output is the pollutant load, metabolic oxygen consumption trend and feeding optimization parameters corresponding to the growth stage. The input and output are deeply bound to the aquaculture growth cycle and water quality stress scenarios.

[0026] The coupled dynamic model, based on real-time feeding behavior and metabolic stress microbehavioral characteristics, matches the nutritional requirements and pollutant production patterns of the target aquaculture species at the corresponding growth stage, and dynamically adjusts core parameters such as feeding frequency, time period, single feeding amount and feed ratio to accurately adapt to the feeding status of aquaculture organisms, reduce the generation of uneaten feed and fecal pollutants from the source, and reduce the initial pollution load of tailwater.

[0027] Based on the dissolved oxygen stress-specific micro-behavioral characteristics of the full-dimensional behavioral fingerprint database, and combined with the real-time dissolved oxygen stress behavior of aquaculture organisms and the metabolic oxygen consumption change trend output by the coupled dynamic model, the trend of dissolved oxygen change in water body is predicted. Multi-level predictive gradient oxygenation is implemented according to the dissolved oxygen risk level, and the number, power and duration of oxygenation equipment are dynamically adjusted according to the biological behavioral response to achieve precise oxygen control and reduce energy consumption.

[0028] According to the above technical solution, the step of implementing graded purification and recycling management of wastewater based on the behavioral characteristics of aquaculture organisms, stress identification results, and prediction results of a coupled dynamic model includes: Based on the coupled dynamic model, combined with the real-time identified feeding behavior characteristics and metabolic stress-related microbehavioral characteristics of aquaculture organisms, the pollutant load range, main pollutant types and output patterns of aquaculture wastewater are predicted, and the wastewater pollution load is divided into three levels: light, medium and heavy; and wastewater purification paths corresponding to different pollution load levels are matched. During the process of injecting purified recycled water into aquaculture, time-series data of the live microbehavior of aquaculture organisms are collected simultaneously. The baseline features and stress-specific features in the full-dimensional behavioral fingerprint database are compared. The safety of the recycled water for aquaculture is determined by using the live behavioral response of aquaculture organisms as the sole benchmark, and the reuse ratio, injection rate and reuse application scenarios of recycled water are dynamically adjusted.

[0029] Based on the same concept as the above embodiments, embodiments of the present invention also provide a graded recycling system for purifying and reusing aquaculture wastewater, the system comprising: Edge-cloud architecture and model building module: This module is used to construct a full-dimensional behavioral fingerprint database of the target aquaculture species. The database includes specific micro-behavioral features corresponding to single-factor water quality stress, dual-factor composite stress, and multi-factor coupled stress, clarifying the one-to-one correspondence between different stress combinations and the subclinical micro-behaviors of aquaculture organisms. Simultaneously, a dynamic model coupling the feeding metabolism, behavioral response, and wastewater pollution load of aquaculture organisms is constructed, along with a multi-stress factor feature decoupling algorithm based on a temporal feature separation network using an attention mechanism. The full-dimensional behavioral fingerprint database, coupled dynamic model, and multi-stress factor feature decoupling algorithm are then lightweighted and compressed, and simultaneously deployed to the cloud and edge gateways to build an edge-cloud isomorphic infrastructure. Multi-source data acquisition and transmission module: used to collect core data on the behavior of aquaculture organisms and auxiliary water quality data, and transmit them to the edge gateway and the cloud; The intelligent decision verification module connects the edge-cloud architecture, model building module, and multi-source data acquisition and transmission module. Based on the full-dimensional behavioral fingerprint database, it uses the micro-behaviors of aquaculture organisms under stress as the sole core basis for determining operating conditions, triggering control measures, and verifying authenticity. First, it filters out pseudo-stresses caused by non-water quality stress. Then, through the multi-stress factor feature decoupling algorithm, it identifies single-factor and compound stress features, stress inducing factors, combination methods, and stress intensity in real stress behavior data. Subsequently, based on the stress identification results, it completes full-gradient standardization judgment of operating conditions. For operating conditions triggering verification, it completes verification through targeted single-variable intervention and changes in biological behavioral responses, outputs control decision instructions, and synchronously updates the full-dimensional behavioral fingerprint database, the multi-stress factor feature decoupling algorithm, and the coupled dynamic model to complete closed-loop iteration. Feeding-oxygenation coordinated control module: Connected to the intelligent decision-making and closed-loop iteration module, it is used to execute feeding-oxygenation coordinated control based on control decision instructions and stress identification results, so as to reduce the pollution load of tailwater from the source; Wastewater grading, purification, and reuse module: Connects the intelligent decision-making and closed-loop iteration module, the edge-cloud architecture and model building module, and is used to perform wastewater grading, purification, and recycling management based on the behavioral characteristics of aquaculture organisms, stress identification results, and the prediction results of coupled dynamic models.

[0030] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0031] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for purifying and reusing aquaculture wastewater through a graded recycling system, applied to a recirculating aquaculture system, characterized in that... include: A comprehensive behavioral fingerprint database of the target aquaculture species is constructed. The database includes specific micro-behavioral features corresponding to single-factor water quality stress, dual-factor composite stress, and multi-factor coupled stress, clarifying the one-to-one correspondence between different stress combinations and the subclinical micro-behaviors of aquaculture organisms. Simultaneously, a dynamic model coupling the feeding metabolism, behavioral response, and wastewater pollution load of aquaculture organisms is constructed, along with a multi-stress factor feature decoupling algorithm based on a temporal feature separation network using an attention mechanism. The comprehensive behavioral fingerprint database, coupled dynamic model, and multi-stress factor feature decoupling algorithm are then lightweighted and compressed, and simultaneously deployed to the cloud and edge gateways to build an edge-cloud isomorphic infrastructure. Collect core data on the live behavior of aquaculture organisms and auxiliary data on water quality, and transmit them to the edge gateway and the cloud; Based on the aforementioned full-dimensional behavioral fingerprint database, and using the micro-behaviors of aquaculture organisms under stress as the sole core basis for determining operating conditions, triggering control measures, and verifying authenticity, the system first filters out pseudo-stresses caused by non-water quality stress. Then, through the aforementioned multi-stress factor feature decoupling algorithm, it identifies the features of single-factor and compound stresses, the type of inducing factors, the combination of stresses, and the intensity of stresses in real stress behavior data. Subsequently, based on the stress identification results, it completes the standardization determination of operating conditions across the entire gradient. For operating conditions that trigger verification, it completes the verification through targeted single-variable intervention and changes in biological behavioral responses, outputs control decision instructions, and simultaneously updates the full-dimensional behavioral fingerprint database, the multi-stress factor feature decoupling algorithm, and the coupled dynamic model to complete the closed-loop iteration. Based on the control decision instructions and stress identification results, a feeding-oxygenation coordinated control is implemented to reduce the pollution load in the effluent from the source; Based on the behavioral characteristics of aquaculture organisms, stress identification results, and prediction results of coupled dynamic models, wastewater is managed through graded purification and recycling.

2. The method for purifying and reusing aquaculture wastewater in a graded, circulating manner according to claim 1, characterized in that, The completion of non-water quality stress pseudo-stress filtration includes: Based on the pre-stored water quality stress-specific micro-behavioral features and non-water quality stress pseudo-stress behavioral features in the full-dimensional behavioral fingerprint database, the system identifies real-time collected live behavior data of aquaculture organisms and extracts the types, durations, and trends of behavioral feature changes. It then links water temperature monitoring data, on-site operation records, stocking density data, and disease-aided monitoring data to match the triggering scenarios and associated causes of behavioral anomalies. The system filters out non-water quality stress pseudo-stress behaviors caused by human disturbance, environmental noise interference, sudden water temperature changes, stocking density stress, and biological diseases, excluding behavioral anomalies caused by non-water quality factors. Finally, it identifies genuine stress behaviors with progressive changes caused by water quality stress and outputs the filtered biological behavioral feature data to subsequent stress decoupling and decision-making processes.

3. The method for purifying and reusing aquaculture wastewater in a graded, circulating manner according to claim 1, characterized in that, The multi-stress factor feature decoupling algorithm of the attention-based temporal feature separation network includes: S1 Input Layer: The time-series data of the micro-behavior of the cultured organisms after pseudo-stress filtering are divided into four types of time-series feature sequences according to the behavioral dimension: respiratory rhythm, swimming characteristics, feeding response, and stress avoidance. After normalization, the data are output to the time-series feature encoding layer. S2 Temporal Feature Encoding Layer: A bidirectional long short-term memory network Bi-LSTM is used to perform contextual encoding on the input multi-dimensional temporal feature sequence, capture the long temporal dependencies and progressive change features of different behavioral features, and output a high-dimensional behavioral feature map that integrates temporal associations. S3 Multi-Scale Attention Separation Layer: A parallel multi-scale channel attention module and a temporal attention module are constructed. The channel attention module learns feature weights corresponding to different water quality stress factors for feature channels of different behavioral dimensions, strengthening stress-related specific feature channels and suppressing irrelevant features. The temporal attention module captures the temporal variation patterns of behavioral features corresponding to different stress factors, locking in key temporal segments of stress features. Through a dual attention mechanism, a weighted output is obtained to obtain decoupled feature subsets corresponding to different stress factors. S4 Decoupling Feature Matching Layer: The decoupled stress feature subsets are matched with the pre-stored single-factor, dual-factor, and multi-factor stress-specific behavioral fingerprints in the full-dimensional behavioral fingerprint database. By using a preset feature similarity threshold, the types and combinations of single-factor stress, dual-factor composite stress, and multi-factor coupled stress are distinguished, and one or more water quality stress triggers that induce stress behavior are accurately located and the corresponding stress intensity is quantified. S5 Output Layer: Outputs the types and combinations of water quality stress factors identified, as well as the corresponding stress intensity, and simultaneously outputs the specific behavioral characteristic data corresponding to each stress factor for subsequent working condition judgment and control.

4. The method for purifying and reusing aquaculture wastewater in a graded, circulating manner according to claim 1, characterized in that, The standardized determination of the full gradient operating condition includes: Pre-set biological behavioral stress grading thresholds for the entire growth cycle of the target aquaculture species, and for single-factor and compound stresses. These thresholds include a slight change threshold corresponding to subclinical mild stress and a stress grading warning threshold corresponding to clear stress abnormalities. The behavioral characteristics of the aquaculture organisms after pseudo-stress filtering and stress decoupling are compared with the stress intensity. The entire process uses biological behavioral characteristics as the sole criterion for judgment, sequentially completing the standardized determination of four types of working conditions: Stress-free steady-state condition: All behavioral characteristics of aquaculture organisms are within the stress-free baseline range. Regardless of fluctuations in water quality auxiliary data, the condition is considered steady-state and no control actions are triggered. Stress-abnormal condition: Behavioral characteristics of aquaculture organisms reach the stress grading threshold. Regardless of whether water quality auxiliary data meets the standards, control actions corresponding to the identified stress triggers are triggered, and the live organism verification process is initiated. Data contradiction condition: Water quality auxiliary data shows anomalies, but the behavioral characteristics of aquaculture organisms remain stable within the stress-free baseline range. The decision-making authority for water quality sensor data is directly locked, and the live organism verification process is initiated. The biological behavioral characteristics are used as the sole criterion for judgment, and control trigger commands corresponding to the three types of conditions are output.

5. The method for purifying and reusing aquaculture wastewater in a graded, circulating manner according to claim 4, characterized in that, The verification process, which involves targeted intervention and changes in biological behavioral responses, includes: Based on the standardized judgment results of the working conditions, for the clearly defined abnormal working conditions that trigger verification, targeted intervention actions are performed on the corresponding water quality factors based on the combination and intensity of water quality stress causes identified by decoupling, and the remaining water quality factors and aquaculture environment parameters are kept stable throughout the process; for the data contradiction working conditions that trigger verification, targeted verification intervention is performed on the corresponding water quality sensors. The entire process involves synchronously collecting time-series data on the microbehavioral behavior of aquaculture organisms before, during, and after the intervention. Data on the magnitude of changes in behavioral characteristics, the rate of mitigation, and the response changes in the duration of these changes are extracted. Using the behavioral response changes of aquaculture organisms as the sole verification benchmark, the changes in behavioral characteristics before and after the intervention are compared to distinguish between actual water quality anomalies and sensor data anomalies. Based on the differentiation and judgment results, corresponding actions are taken: For results determined to be true water quality anomalies, emergency control measures matching the stress type and intensity are implemented, and the characteristics of the abnormal behavior and intervention response data are simultaneously added to the full-dimensional behavioral fingerprint database; for results determined to be sensor drift distortion, the time point at which the stress behavior of aquatic organisms completely returns to the stress-free baseline range after targeted intervention is identified, and the online correction of sensor parameters is completed based on the measured data collected synchronously at that point; for results determined to be sensor fault jumps, equipment self-checks and fault alarms are triggered, and the control decision-making authority for abnormal data is locked.

6. The method for purifying and reusing aquaculture wastewater in a graded, circulating manner according to claim 1, characterized in that, The aforementioned control decision-making instructions and stress identification results are used to execute a coordinated feeding-oxygenation control system to reduce effluent pollution load at the source, including: Based on the coupled dynamic model, combined with the real-time identified feeding behavior characteristics and metabolic stress microbehavioral characteristics of farmed organisms, the nutritional requirements and pollutant production patterns of the target farmed species throughout their entire growth cycle are matched, and the feeding plan and core feeding parameters are dynamically adjusted to reduce the generation of uneaten feed and fecal pollutants from the source. Based on the dissolved oxygen stress-specific micro-behavioral features pre-stored in the full-dimensional behavioral fingerprint database, combined with the real-time identified dissolved oxygen stress behavior features of aquaculture organisms and the metabolic oxygen consumption change trend output by the coupled dynamic model, the trend of dissolved oxygen change in water body is predicted, multi-level predictive gradient oxygenation actions are executed, and the operating parameters of oxygenation equipment are dynamically adjusted.

7. The method for purifying and reusing aquaculture wastewater in a graded, circulating manner according to claim 1, characterized in that, The process of implementing graded purification and recycling control of wastewater based on the behavioral characteristics of aquaculture organisms, stress identification results, and prediction results from a coupled dynamic model includes: Based on the coupled dynamic model, combined with the real-time identified feeding behavior characteristics and metabolic stress-related microbehavioral characteristics of aquaculture organisms, the pollutant load range, main pollutant types and output patterns of aquaculture wastewater are predicted, and the wastewater pollution load is divided into three levels: light, medium and heavy; and wastewater purification paths corresponding to different pollution load levels are matched. During the process of injecting purified recycled water into aquaculture, time-series data of the live microbehavior of aquaculture organisms are collected simultaneously. The baseline features and stress-specific features in the full-dimensional behavioral fingerprint database are compared. The safety of the recycled water for aquaculture is determined by using the live behavioral response of aquaculture organisms as the sole benchmark, and the reuse ratio, injection rate and reuse application scenarios of recycled water are dynamically adjusted.

8. A graded, recirculating aquaculture wastewater purification and reuse system, characterized in that: The system includes: Edge-cloud architecture and model building module: This module is used to construct a full-dimensional behavioral fingerprint database of the target aquaculture species. The database includes specific micro-behavioral features corresponding to single-factor water quality stress, dual-factor composite stress, and multi-factor coupled stress, clarifying the one-to-one correspondence between different stress combinations and the subclinical micro-behaviors of aquaculture organisms. Simultaneously, a dynamic model coupling the feeding metabolism, behavioral response, and wastewater pollution load of aquaculture organisms is constructed, along with a multi-stress factor feature decoupling algorithm based on a temporal feature separation network using an attention mechanism. The full-dimensional behavioral fingerprint database, coupled dynamic model, and multi-stress factor feature decoupling algorithm are then lightweighted and compressed, and simultaneously deployed to the cloud and edge gateways to build an edge-cloud isomorphic infrastructure. Multi-source data acquisition and transmission module: used to collect core data on the behavior of aquaculture organisms and auxiliary water quality data, and transmit them to the edge gateway and the cloud; The intelligent decision verification module connects the edge-cloud architecture, model building module, and multi-source data acquisition and transmission module. Based on the full-dimensional behavioral fingerprint database, it uses the micro-behaviors of aquaculture organisms under stress as the sole core basis for determining operating conditions, triggering control measures, and verifying authenticity. First, it filters out pseudo-stresses caused by non-water quality stress. Then, through the multi-stress factor feature decoupling algorithm, it identifies single-factor and compound stress features, stress inducing factors, combination methods, and stress intensity in real stress behavior data. Subsequently, based on the stress identification results, it completes full-gradient standardization judgment of operating conditions. For operating conditions triggering verification, it completes verification through targeted single-variable intervention and changes in biological behavioral responses, outputs control decision instructions, and synchronously updates the full-dimensional behavioral fingerprint database, the multi-stress factor feature decoupling algorithm, and the coupled dynamic model to complete closed-loop iteration. Feeding-oxygenation coordinated control module: Connected to the intelligent decision-making and closed-loop iteration module, it is used to execute feeding-oxygenation coordinated control based on control decision instructions and stress identification results, so as to reduce the pollution load of tailwater from the source; Wastewater grading, purification, and reuse module: Connects the intelligent decision-making and closed-loop iteration module, the edge-cloud architecture and model building module, and is used to perform wastewater grading, purification, and recycling management based on the behavioral characteristics of aquaculture organisms, stress identification results, and the prediction results of coupled dynamic models.

9. A computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the steps of the method as claimed in any one of claims 1 to 7.

Images